Method for interpreting kidney ultrasound images with artificial intelligence

Abstract

A method for interpreting kidney ultrasound images with artificial intelligence includes: using a deep-learning method on abnormal and non-anomalous kidney ultrasound images obtained from classification by experienced nephrological experts and via image pre-processing to obtain prediction modules, and upon or after obtaining the prediction modules, obtaining a rules module establishing specific rules for combination and sequence among the prediction modules, and obtaining an interpretation model from a combination of the prediction modules and the rules module to predict whether a kidney ultrasound image is abnormal; wherein each prediction module is used to determine the probability of predefined abnormal pattern, and the rules module provides the logic for determining the prediction modules. Accordingly, the method for interpreting kidney ultrasound images with artificial intelligence is able to allow non-kidney image interpretation experts to find kidney abnormality early and to use it as the basis for future remote medical care item.

What is claimed is: 
     
       1. A method for interpreting kidney ultrasound images with artificial intelligence, comprising the following steps:
 Step a: using a computer to obtain a plurality of pre-processed kidney ultrasound images from a database in order to establish a plurality of first training image sets and a second training image set; wherein the plurality of first training image sets correspond to different classifications of abnormal patterns respectively, and each one of the plurality of training image sets comprises a plurality of kidney ultrasound images belonging to one of the abnormal patterns classified, and the second training image set comprises a plurality of non-anomalous children kidney ultrasound images classified; 
 Step b: using at least one of the kidney ultrasound images in the first training image sets and at least one of the kidney ultrasound images in the second training set as a training data, using a deep-learning training model to perform classification and computation in order to obtain an interpretation model, wherein: 
 the interpretation model comprises a plurality of prediction modules corresponding to the abnormal patterns respectively and a rules module; wherein the rules module is a model analysis logic obtained from a result of abnormal probability and normal probability obtained from comprehensive analysis of the prediction modules, in order to select all or a portion of the prediction modules and to form at least two sets of different interpretation sequence combinations; 
 Step c: using the computer to perform analysis on a kidney ultrasound image to be analyzed pre-processed by the computer according to the rules module and based on the selected interpretation sequence via all or a portion of the prediction modules, in order to obtain a prediction classification result of the kidney ultrasound image to be analyzed; wherein the prediction classification result is used to understand the kidney ultrasound image to be analyzed belongs to an abnormal classification or a non-anomalous classification; 
 wherein the pre-process undergone by the kidney ultrasound images and the children kidney ultrasound image to be analyzed further comprises cleaning of an auxiliary information and capturing of a region of interest; the auxiliary information comprises a text, an auxiliary pattern, a symbol or a combination of at least two thereof; at least a portion of the region of interest comprises an image corresponding to a kidney or further comprises an image of a portion of other human tissue adjacent to the kidney. 
 
     
     
       2. The method for interpreting kidney ultrasound images with artificial intelligence according to  claim 1 , wherein the abnormal patterns comprise a hydronephrosis, a large kidney cyst, a small kidney cyst, a large kidney stone, a small kidney stone, a hyperechoic kidney or a combination of at least two thereof. 
     
     
       3. The method for interpreting kidney ultrasound images with artificial intelligence according to  claim 1 , wherein in Step b, the deep-learning model adopts a transfer learning to perform training. 
     
     
       4. The method for interpreting kidney ultrasound images with artificial intelligence according to  claim 1 , wherein in Step b, according to a predefined weight, the plurality of kidney ultrasound images complying with a quantity of the predefined weight are respectively obtained from each one of the first training image sets and the second training image set as the training data, and the remaining kidney ultrasound images not selected as the training data are used as a test data. 
     
     
       5. The method for interpreting kidney ultrasound images with artificial intelligence according to  claim 1 , wherein each one of the first training image sets is established according to a predefined abnormal classification determination standard. 
     
     
       6. A system for interpreting kidney ultrasound images with artificial intelligence, comprising:
 a processing unit having a processing member for receiving a kidney ultrasound original image to be analyzed, and after performing an image pre-processing procedure, using the interpretation model according to  claim 1  to perform analysis and to obtain a prediction result, in order to understand a classification of an abnormal pattern or a non-anomalous pattern of a kidney and an accuracy rate thereof; and 
 an ultrasound machine having an image generation member providing the kidney ultrasound original image to be analyzed, an image transfer module for transferring the kidney ultrasound original image to be analyzed to the processing member. 
 
     
     
       7. The system for interpreting kidney ultrasound images with artificial intelligence according to  claim 6 , wherein the processing unit, the ultrasound machine and the display unit are connected to each other via a wired or a wireless method.